Data Analysis and Results

The first hypothesis is addressed by looking at the descriptive statistics. Shares of applications and admissions of rural and urban⁶ students were calculated and grouped by the field of study.

Table C2 reveals unequal patterns in educational choice between the two groups. The largest differences are observed within ICT, Agriculture and Education. ICT, as the most lucrative field, shows the prevalence of urban applicants - 12.7% of urban schools graduates submitted applications to ICT programs and 8.6% were admitted, by contrast to 7.3% and 4.9%

respectively among rural graduates. Meanwhile, the share of rural students entering low

6 “Urban” category here includes urban settlements disregarding size and type.

CEUeTDCollection

26

prestige fields of Agriculture and Education is two times larger than among their metropolitan peers. The pattern of higher education choices and admissions are more similar among urban students, grouped by the size of the city. The biggest gap again occurring in the field of ICT (Table C3). This finding is consistent when comparing applicants from the regular and elite schools - the second group outperforms the first in terms application and admission to ICT by 4%. Similar to the comparison of rural and urban applicants, the largest gap is observed for Education. 17.35% of students from regular schools entered this field compared to 12.2%

among elite schools graduates. Looking at the dimension of university choice, as expected, elite schools graduates more often enter the most prestigious universities. 63.4% of them gained admissions from the highest selectivity rank compared to 51% of applicants coming from regular schools. Overall, the descriptive analysis has supported the hypothesis 1 regarding differences in patterns of applications and admissions among groups of students conditional on socioeconomic profiles. To further explore this relationship, the analysis proceeds with the model of application behavior.

3.2.1. Model of Applying to the Prestigious Universities and Fields of Studies

The binary logistic regression model was fitted on two outcomes - applying to a prestigious university and a field of study. Table 3.4 reports the average odds ratios. The reference groups are students from large cities and capital (residential location variable), and graduates of the elite schools (type of school variable). Thus, the odds ratio values less than one indicate lower probabilities of applying for students of less advantageous socioeconomic origin.

In line with the theoretical expectations, for both outcome variables, residential location and type of school appear to be significant predictors of submitting the ambitious application.

Students of non-selective secondary schools are about 11% less likely to apply to prestigious universities and fields of study, holding other factors constant. Among residential location

CEUeTDCollection

27

categories, both models estimated the lowest odds of competing for the selective degrees for the rural applicants. Their odds of submitting an applications are by almost a half lower than among urban students.

Consistent with the findings of the similar studies (Davies and Guppy, 1997; Chankseliani, 2013), controlling for academic achievements (Table 3.3, column 6; models with and without academic achievements controls are provided in the Table B1) increases the odds associated with applying to a selective university for applicants from disadvantaged backgrounds.

However, the odds ratio remains significant and lower than one, indicating that, even in the hypothetical situation of achieving the same scores as more advantageous peers, students from smaller settlements are less likely to submit applications to the prestigious universities. The closest to one (or absence of differences) is the group of regular school students, compared to elite schools graduates, holding else equal. It is important to acknowledge that, students from regular school are expected to be the most diverse group in this analysis in terms of socioeconomic status since public education is prevelant and it engages not only students of lower and middle classes.

Interestingly, in the model of application to the prestigious fields of study, the odds of socioeconomic indicators remain almost unaltered for students from three residential categories with the addition of academic achievements and compensatory coefficients controls (Table 3.3, column 2). The possible explanation is context-specific. Higher education expansion with no quality assurance system has led to the situation when credentials became more important than quality and relevance of acquired skills (World Bank 2019, 10). Therefore, Ukrainian students and their families may perceive receiving the degree from the prestigious university as more important factor of future success than enrolling in the program in the certain field, so the academic competition is higher for places in universities than within the fields. This finding is also aligned with the literature on educational choice, suggesting that , when the choice of

CEUeTDCollection

28

institution is influenced by the composition and hierarchy of educational system (Boliver 2015), as well as determined by individual test scores, the choice of a field is more likely to derive from individual prior experiences, abilities and preferences (Mcmaster 2019). The test for goodness of fit reveals supports this argument - the model of application to a prestigious university fits data better (McFadden R² = 0.10) than the model of application to the selective fields (McFadden R² = 0.04), indicating that available socioeconomic and application-specific indicators do not fully explain the variations in educational choices.

CEUeTDCollection

29

Regional Coefficient -1.073^*** 0.342^*** 0.057^*** 1.059^***

(0.018) (0.018) (0.013) (0.013)

Subject-Specific Coefficient

Regional Coefficient -0.602^*** 0.548^*** -3.910^*** 0.020

(0.012) (0.012) (0.042) (0.042)

Constant -4.532^*** 0.011 -1.047^*** 0.351^***

(0.039) (0.039) (0.030) (0.030)

Year Yes Yes Yes Yes

Observations 4,348,111 4,348,111 4,348,111 4,348,111

Log Likelihood -2,620,746.0 -2,620,746.0 -2,849,857.0 -2,849,857.0 Akaike Inf. Crit. 5,241,518.0 5,241,518.0 5,699,741.0 5,699,741.0

*p<0.1; ^**p<0.05; ^***p<0.01

CEUeTDCollection

30

Note: Table reports coefficients and odds ratios. Standard errors are clustered at the school level. The number of observations dropped due to missing values in the school type variable.

3.2.2. Model of Admission to a Prestigious University and Field of Study

The analysis proceeds with modeling the admission outcome. Using the same analytical setup of binary logistic regression, the model explores the relationship between students’

socioeconomic characteristics and enrollment to the institutions and fields of study of the highest prestige (Tables 3.4). The admission models include a new control variable – binary indicator of whether the application was submitted as the first choice. It is a significant predictor of admission since the automated system of sorting of applicants adjusts the number state-sponsored places considering the number of entrants with the high test scores and indicated first choice in the application list.

Overall, the odds of admission to a prestigious university and a field of study are larger than odds of applying, suggesting self-selection. It is a likely scenario, considering that the electronic admission system allows students to see scores and positions of other applicants in the list, and evaluate own chances before the submission.

Consistent with the hypothesis 2, students’ residential location is significantly and meaningfully associated with odds of admission to a selective university. In particular, the odds of students from the five largest cities are 1.32 times higher for gaining admission in a prestigious university, compared to rural applicants. In line with the expectations, the odds are closer to one for students from periphery cities and regional centers. As was observed in the application models, the negative effect of attending regular school diminishes once academic achievements, residential status and compensatory coefficients variables are controlled.

Since the main goal of the compensatory coefficients is to mitigate the systematic disparities in academic achievements of students and provide applicants with chance to get a place in the

CEUeTDCollection

31

university, it is important to look at their effect for the odds of gaining admission, when academic achievements are not included in the model (Tables B3 and B4). The associated coefficients are statistically significant and directions of coefficients is as expected: subject-specific measure negatively effects the probability of admission to a prestigious field, regional – to a prestigious university. However, the inclusion of these variables into the model does not result in notable changes in odds ratio values.

In a summary, both sets of models suggested statistically significant and positive relationship between residential location and odds of applying or gaining an admission in a prestigious university or field of study. The probabilities of entering a selective higher education is the lowest for rural students, the most disadvantaged group in terms of prior educational opportunities. The negative effect of residential location does not disappear even when academic achievements are controlled. It suggests that, in addition to lower test scores, students outside of large metropolitan areas, especially in rural settlements, experience other factors of diversion from the first-tier schools and programs. The attendance of regular school is also associated with lower odds of gaining admission, and, the gap is diminishing once academic achievements are controlled.

The models also showed the evidence of double stratification effect of socioeconomic origin.

Smaller settlement size and attendance of non-selective schools results in lower odds of getting into prestigious higher education in regards to both fields of study and universities. Although the associated coefficients for compensatory measures are significant, the analysis did not reveal the evidence of a meaningful stratification effects for these variables.

CEUeTDCollection

32

Table 3.4. Binary Logistic Model of Admission to a Prestigious University and Field of Study Model of Admission

Dependent variable:

Prestigious University Prestigious Field Coefficients Odds ratio Coefficients Odds ratio

Rural -0.375^*** 0.687^*** -0.317^*** 0.728^***

Regional Coefficient -0.129^*** 0.879^*** 0.278^*** 1.320^***

(0.013) (0.013) (0.014) (0.014)

Subject-Specific Coefficient 0.215^*** 1.240^*** -3.045^*** 0.048

(0.014) (0.014) (0.060) (0.060)

Observations 4,348,111 4,348,111 4,348,111 4,348,111

Log Likelihood -733,347.30 -733,347.300 -674,952.30 -674,952.30 Akaike Inf. Crit. 1,466,723.0 1,466,723.000 1,349,933.0 1,349,933.0 Note: Table reports coefficients and odds ratios. Standard errors are clustered at the school level. The number of observations dropped due to missing values in the first choice variable.

CEUeTDCollection